Using Machine learning to enable feedback controlled manufacture of self-assembled patterned materials

在小长度尺度上实现对材料结构的控制是实现新技术的关键要求，这些新技术在能源生产、能源储存和医疗保健领域具有更好的性能。举一些例子，太阳能电池，电池和医疗诊断传感器都可以受益于纳米和微米尺度的结构。制造这种结构的一种有吸引力的方法是通过自组装，其中材料将自己排列成定义明确的规则图案。获得这种行为的一种方法是通过干燥含有感兴趣的材料的悬浮液或溶液。例如，如果将颗粒在液体中的悬浮液铺展到固体载体上并干燥，在某些条件下将产生非常规则的重复“结晶”排列。这些排列具有可用于传感器的光学特性，并可作为模板来制造电池和太阳能电池材料的有效电极。在干燥溶液中有用的结构形成过程的其他实例包括由多组分混合物形成的复杂网络和规则晶体结构，如果以某些特定长度尺度存在，这两者都可以增强太阳能电池的性能。然而，尽管其简单性和潜在的应用，目前在干燥溶液中的自组装主要被用作生产少量材料的研究方法，而不是被视为常规的制造方法。我们认为，这是因为自组装对许多参数高度敏感，这阻碍了可重复性，并且需要耗时的优化来生产给定的材料，使其对于大规模制造没有吸引力。在这里，我们计划调查，如果添加一个自动化控制系统，以连续自组装/解决方案结构为基础的过程可以克服这些障碍。为了实现控制系统，我们将在自组装过程中进行微观观察，并根据此反馈使用算法来调整制造仪器参数。该方法将用于快速确定生产特定应用的理想结构所需的参数，以及在连续制造大量材料期间保持高质量的均匀结构生产。尽管反馈被很好地确立为控制其他领域中的制造过程的有效方式，但是由于观察结构形成过程的困难以及实现所需算法以有利地调整参数的挑战，该方法尚未被部署用于自组装。在这里，我们将利用干燥薄膜的真实的实时监测方面的进展，以及能够为复杂行为建立模型的“机器学习”计算机方法的专业知识，来克服这些挑战。

Achieving control over materials' structure at small length scales is a critical requirement to realise new technologies with improved performance in areas of energy generation, energy storage and healthcare. To give some examples, solar cells, batteries and sensors for medical diagnostics can all benefit from nano and micron scale structuring. An attractive way to produce such structures is via self-assembly, where materials arrange themselves into well-defined regular patterns. One way to access this behaviour is by drying a suspension or solution containing the material of interest. For example, if a suspension of particles in a liquid is spread onto a solid support and left to dry, under some conditions very regular, repeating "crystalline" arrangements will result. These arrangements possess optical properties that can be used for sensors, and serve as templates to make efficient electrodes for batteries and solar cell materials. Other examples of useful structure formation processes in drying solutions include complex networks that form from multi-component mixtures, and regular crystalline structures, both of which can enhance the performance of solar cells if present at certain specific length scales. However, despite its simplicity and potential applications, at present self-assembly in drying solutions is mainly used as a research method to produce small quantities of material, and is not viewed as a routine manufacturing method. We believe that this is because self-assembly is highly sensitive to many parameters, which hampers reproducibility and requires time consuming optimisation to produce a given material, making it un-attractive for large scale manufacturing. Here, we plan to investigate if adding an automated control system to continuous self-assembly/solution structure based processes can overcome these obstacles. To implement the control system, we will make microscopic observations during self-assembly and use algorithms to adjust the manufacturing instruments parameters based on this feedback. This method will be used both to rapidly identify the parameters required to produce the ideal structure for a particular application, and also to maintain high quality uniform structure production during continuous manufacture of large amounts of material. Despite feedback being well established as an effective way to control manufacturing processes in other sectors, this method has not yet been deployed for self-assembly due to difficulties in observing the structure forming process, and the challenges of implementing the required algorithms to beneficially adjust parameters. Here we will use advances in real time monitoring of drying films, together with expertise in "machine learning" computer methods that are able to build models for complex behaviour, to overcome these challenges.